The question arising from these data was of the functional significance of the colocalization of TEAD and RUNX1 with AP-1. TEAD is a major mediator of HIPPO signaling, which together with its co-factors YAP/TAZ is absolutely essential for the EHT (Goode et al., 2016) where it is required to switch on RUNX1 in response to the onset of blood flow (Lundin et al., 2020). The integration of ChIP data for TEAD and peaks bound by JUN/FOS dimers in FLK1+ cells representing the differentiation stage prior to the HE, showed a significant overlap between the binding sites for the two factors in the HE (Obier et al., 2016; Goode et al., 2016), in particular at vascular genes whose expression was significantly reduced after dnFOS expression. We therefore performed ChIP experiments for TEAD4 before and after dnFOS expression (Figure 3A) and demonstrated that the abrogation of AP-1 binding abolished TEAD binding as well, demonstrating a close collaboration of the two factors in defining gene expression in the hemogenic endothelium and thus integrating several independent signaling pathways (Obier et al., 2016) (a re-analysis of this data is shown in Figure 3B). We have not measured AP-1 binding in the absence of TEAD, but it is likely that the formation of endothelial cells including the HE involves AP-1 dependent TEAD4 de novo binding, suggesting a cooperative requirement role for AP-1. Our findings in primary cells adds to the increasing number of studies which described a growth-promoting role of AP-1/TEAD interactions in cancer cell lines demonstrating that this type of signaling integration is wide-spread across multiple tissues (Liu et al., 2016; Zanconato et al., 2015; Koo et al., 2020). However, during hematopoietic specification, AP-1 and HIPPO signaling also play an instructive role as both are required for the establishment (Lundin et al., 2020) and maintenance of the endothelial cell fate, as in the absence of AP-1 the balance is shifted towards hematopoietic cells (Obier et al., 2016).

The role of the co-localization of RUNX1 and AP-1 was more difficult to explain. The AP-1 motif was found in both HE1-specific and HP-specific DHSs, but not at dnFOS unresponsive sites, which are dominated by RUNX1 and ETS motifs (Figure 3C, TF motif plots), highlighting a different function of these binding sites in endothelial and hematopoietic cells. RUNX1 is absolutely required for the EHT (Lancrin et al., 2009) and the activation of the hematopoietic gene expression program. Importantly, RUNX1 reorganizes the chromatin landscape by relocating other TFs such as SCL/TAL1, LDB1 and FLI-1 to new binding sites (Lichtinger et al., 2012; Gilmour et al., 2018). It has also been shown previously in T cells that RUNX1 binding can be dependent on the NFAT/AP-1 complex (Johnson et al., 2004). We therefore performed RUNX1 ChIP experiments in the presence and absence of dnFOS in HP cells (Figure 3C). As seen with TEAD, also here the expression of dnFOS leads to a loss of at least 1,800 distal open chromatin regions, but also with a gain of about 1,000 new binding sites (Figure 3C). TF binding motif analysis of lost and gained regions showed a profound loss of binding motifs for RUNX1 and AP-1 (as expected) in addition to those of the myeloid TFs PU.1 and C/EBP family members, providing support for the finding that AP-1 is essential for myelopoiesis (Obier et al., 2016) and adds that the cooperation of AP-1 and RUNX1 is essential for this process. Interestingly, inspection of the genes associated with lost RUNX1 sites (Supplemental dataset 1) revealed numerous transcriptional regulators, including JUN and FOS themselves suggesting a complex feedback relationship between RUNX1 and the AP-1 factor family (shown for Fos in Figure 3D). Gained sites show enrichment of RUNX, GATA and ETS motifs, indicating that RUNX1 moves towards such sites which are characteristic for a stem cell/erythroid fate, again demonstrating that the cooperation of AP-1 with tissue-specific factors determines cellular fates. In summary, our data demonstrate that for the first step in differentiation to occur, such a mechanism does not require to activate a new set of regulators, it only requires to relocate cooperating pre-existing and inducible factors to new cis-regulatory elements and then activate a new set of genes. The question now arose of the nature of the cis-regulatory elements upon which such cooperation takes place.

The developmental control of gene expression critically depends on enhancer elements which interact with promoters to activate gene expression and define the dynamics of expression of a given gene in a specific regulatory context, and the hematopoietic system is not exception (Field and Adelman, 2020; Mulet-Lazaro and Delwel, 2023). Multiple surrogate assays based of chromatin structure and histone modifications have been used to measure the number of enhancer elements in the genome. However, the definition of an enhancer element is functional, i.e., the ability to stimulate the transcriptional activity of a promoter (reviewed in (Field and Adelman, 2020; Mulet-Lazaro and Delwel, 2023)). In order to functionally identify which CREs had the ability to stimulate reporter gene expression in a chromatin context during hematopoietic specification, we developed a high-throughput method based on isolated ATAC-Seq fragments that identified thousands of differentially active CREs able to stimulate a minimal promoter integrated within a safe harbor site within the genome of mESCs (Edginton-White et al., 2023). This system enables to measure enhancer and promoter activity at each stage of differentiation into blood and identified thousands of differentially active CREs. Enhancer elements could also be tested individually to examine the role of specific TF binding motifs. Once defined as functional enhancers, we could then test (i) at which developmental stage such elements were visible as an open chromatin site, (ii) at which stage they develop enhancer activity and (iii) whether such elements are responsive to extrinsic signals and when/how (Edginton-White et al., 2023; Maytum et al., 2024).

Several results were noteworthy: (i) After we filtered out repeat and promoter elements, the majority of distal open chromatin regions showed stimulatory activity in our assay; (ii) the presence of open chromatin did not always indicate that this element was capable to reading out in the enhancer assay at this developmental stage which we defined as a primed state, and (iii) thousands of open chromatin regions were dependent on extrinsic signals which we showed by omitting specific growth factors from the differentiation cultures. Open chromatin was measured in sorted cells making sure that differences in cell type composition were accounted for (Figure 4). We measured the effects of 4 different cytokines: BMP4, which is important to pattern mesoderm, vascular endothelial growth factor (VEGF) which is required to generate endothelial cells, including the hemogenic endothelium (Ottersbach, 2019; Maytum et al., 2023) and Interleukin 3 and 6 both of which are required for the growth of multiple blood lineages (Broughton et al., 2012; Hirayama and Ogawa, 1996) (Figure 4A). Moreover, (iv) when we constructed GRNs from this data, we observed that a subset of connections within the GRN of one developmental stage anticipated the next state, with specific enhancers being organized in open chromatin prior to the onset of gene expression at the next stage, demonstrating that chromatin opening follows a defined trajectory. Some of these sites were also signaling responsive, suggesting a role of cytokines in chromatin priming (Maytum et al., 2024). Finally, (v) co-localization analysis showed that AP-1 motifs were ubiquitously present, sitting next to different cell-type specific factors at each differentiation stage, with AP-1-TEAD pairs being prominent at the endothelial stage and AP-1 shifting its alliance to hematopoietic TFs as seen before. The ChIP assays confirmed this notion. Figure 5 illustrates that the vast majority of dnFOS responsive TEAD4 and RUNX1 binding sites are localized in functionally active enhancers (Figure 5A), with AP-1-TEAD pairs being prominent at the endothelial stage and AP-1 shifting its alliance to RUNX1 at the hematopoietic progenitor stage (Figure 5B).

To understand the molecular basis of cytokine responsiveness of enhancer elements we examined one signaling pathway (VEGF signaling) in more detail (Edginton-White et al., 2023). Strikingly, more that 8,000 enhancers were responsive to the presence or absence of VEGF across all differentiation stages, i.e., showed an at least two-fold change in chromatin accessibility. A careful analysis of the TF binding motifs underlying responsiveness showed that the presence of VEGF was associated with the activation of CREs carrying AP-1 and TEAD motifs, in concordance with the finding that VEGF signals via AP-1 (Kellaway et al., 2024; Jia et al., 2016). The absence of VEGF was associated with the enrichment of RUNX1 motifs, suggesting that VEGF suppresses the activation of such enhancers. Indeed, timed withdrawal experiments demonstrated that hematopoietic development was strongly suppressed by VEGF. Moreover, the presence of VEGF is absolutely required for the establishment of non-hemogenic endothelium but then the maintained signaling input from VEGF suppressed the activation of enhancer elements important for the expression of Runx1, thus blocking the full execution of the EHT which requires RUNX1 upregulation. In addition, the removal of VEGF upregulated a repressor of NOTCH-signaling which needs to be down-regulated for the EHT to occur. To study the interaction of VEGF signaling responsive AP-1 factors and TEAD factors we interrogated one enhancer (for the Galnt1 gene) which we identified as being active in the HE and repressed after the EHT which contained both AP-1 and TEAD binding motifs and which we confirmed by ChIP-Seq was bound by TEAD4, FOS, JUN in the HE and then RUNX1 in the HP (Edginton-White et al., 2023). It turned out that its activity was regulated by a composite AP-1/TEAD element with a 7 base pair space bioinformatically identified in (Obier et al., 2016) which overlapped with a RUNX1 site, thus suggesting that the balance between the three factors determined whether an element was activated after the EHT or not.

To examine this result at the global level we integrated our dnFOS data with the enhancer data and found approximately one thousand enhancers bound by FOS. Only about 50% of these elements were still active in HP cells indicating that half of FOS bound enhancer sites were lost, again showing the cell-type specificity of AP-1 binding. Very few enhancers bound by FOS in the HE were only active in HP cells. To test whether dnFOS responsive TEAD and RUNX1 binding sites were also responsive to VEGF signaling, we integrated these data with ATAC-data from HE and HP cells formed in the presence and absence of VEGF. We found that 44% of all dnFOS responsive TEAD4 binding sites and and 52% of dnFOS-responsive RUNX1 binding sites were also VEGF responsive, confirming that the cooperation between AP-1 and these factors is driven by signaling. The conclusion from these experiments is therefore that the presence or absence of cytokines has a profound and cell-type specific influence on the chromatin landscape and on cell differentiation via the differential activity of enhancer elements.

Both the JUN and the FOS genes were originally isolated from oncogenic retroviruses suggesting that their mis-expression could transform cells (Ransone and Verma, 1990). The fact that cell cycle progression and AP-1 activity are closely linked has long been recognized as an essential part of the cellular transformation process (Shaulian and Karin, 2002; Angel and Karin, 1991) as AP-1 inhibition blocks oncogenesis in multiple contexts (Olive et al., 1997; Tichelaar et al., 2010). This is also true for Acute Myeloid Leukaemia (AML). The analysis of the GRNs maintaining mutation-specific AML sub-types revealed that AP-1 binding sites are a prominent node in the GRN of all studied sub-types. Moreover, employing the dnFOS peptide, we could show that blocking AP-1 DNA binding activity blocked the proliferation of several AML sub-types in vitro and in vivo (Martinez-Soria et al., 2019; Kellaway et al., 2024; Assi et al., 2019; Adamo et al., 2023; Coleman et al., 2023). The same was true for RUNX1. Inhibition of the DNA binding activity of this TF using a small molecule inhibitor (Illendula et al., 2016) or down-regulating its expression blocked the growth of multiple AML sub-types as well (Adamo et al., 2023; Coleman et al., 2023; Ben-Ami et al., 2013; Wesely et al., 2020). For two AML sub-types, the t (8; 21) translocation which expresses an aberrant RUNX1 fusion protein, RUNX1-ETO and the FLT3-ITD which expresses a constitutively active receptor for the cytokine FLT3-ligand, we were able to gain insights into the molecular mechanism of RUNX1 action, the signals activating AP-1 and its genomic targets involved in driving cell proliferation. As it turned out, the signals were highly heterogeneous. The molecular basis of driving proliferation was not.

In recent years, several inhibitors of FLT3 signaling were developed that provided temporary clinical benefit for patients, but most of them eventually relapsed (Perl et al., 2022). To gain insight into why this was the case, we generated GRNs of cells from patients before FLT3 inhibitor (FLT3i) treatment and after relapse (Coleman et al., 2024). We also studied a patient that was unresponsive to the drug. The comparison of the AML-specific GRN before and after relapse showed that the biggest changes were seen in the AP-1 and the RUNX1 nodes. AP-1 mediated connections were upregulated, whilst RUNX1 connections were down-regulated. The unresponsive patient showed no or little change in their GRN. To test, how FLT3i affected AP-1 binding in responsive cells that stopped proliferating after inhibitor treatment, we performed ChIP assays which demonstrated that both AP-1 and RUNX1 binding to chromatin was down-regulated, demonstrating that AP-1 was the main mediator of FLT3 signaling with the expression of cell cycle genes being strongly affected. Interestingly, the down-regulation of binding occurred at sites with co-localizing AP-1/RUNX1 motifs and as seen during hematopoietic specification (Figure 3), experiments using dnFOS showed that RUNX1 binding at these sites was indeed dependent on AP-1 binding. Moreover, the small molecule inhibition of RUNX1 also led to a profound cell cycle block (Coleman et al., 2023). The explanation as to why RUNX1 connections were lost after relapse was somewhat unexpected. FLT3i treatment and the cell cycle block led to a strong upregulation of multiple signaling genes, such as KIT, all of which are targets of RUNX1 which primed the cells for bypassing inhibition of the FLT3 signal by using signals from other cytokines. Indeed, the addition of cytokines such as IL-3 immediately restored growth and RUNX1 binding. In relapse patients, RUNX1 was still required for growth, but its GRN was rewired towards a lesser dependence on RUNX1.

The precise molecular mechanism by which the AP-1/RUNX1 axis drives growth in t(8;21) cells was completely different (Kellaway et al., 2024). Here, the driver oncogene, RUNX1-ETO directly influences the cell cycle by counteracting the role of normal RUNX1 (Martinez-Soria et al., 2019; Nafria et al., 2020). However, the cells strike a fine balance because wild-type allele of RUNX1 is still present and, as in other AML types, is required for growth (Ben-Ami et al., 2013). To develop rapidly growing blast cells expressing RUNX1-ETO, cells therefore acquire activating mutations in signaling pathways such as those generating a constitutively active RAS or KIT (Krauth et al., 2014). However, this type of bypass only operates in blast cells, as it was shown that the activation of such genes in HSCs on their own leads to stem cell exhaustion (Di Genua et al., 2019). In leukemic stem cells (LSCs) which are mostly quiescent, the IL-5 and VEGF pathways are activated which both signal to AP-1 and in HSCs still are organized in a chromatin structure that is permissive for activation (Kellaway et al., 2024). Once these pathways are active, LSCs start to grow and differentiate into leukemic blast cells. dnFOS expression leads to a loss of RUNX1 binding in chromatin, again showing that proliferation is regulated by the RUNX1/AP-1 axis. The combination of signaling mediated AP-1 activity and AP-1-dependent RUNX1 binding therefore generates a feed-forward loop that kick-starts leukemic growth.

The examples described above give a brief glimpse into the breath-taking complexity of mechanisms and pathways of how normal and malignant cells use AP-1 to integrate multiple extrinsic signals in a chromatin environment to control cell fate via the activity of thousands of enhancer elements and cell proliferation via the direct regulation of cell cycle and signaling genes. JUN/FOS are relatively small proteins with a simple structure (Figure 2B). The only discernable domain is the leucine-zipper domain, whilst the rest of the protein and its transactivation domain are intrinsically disordered (Campbell and Lumb, 2002) (Figure 2B). How do they do it?

The answer, as already alluded to above, lies in the intrinsic ability of these proteins to use their flexible trans-activation domain to interact with multiple other TFs. Such interactions can be guided by DNA on composite binding sites, as exemplified by the AP-1/NFAT complex (Johnson et al., 2004; Chen et al., 1998), by interdependent binding of juxtaposed individual binding sites as seen with TEAD, or by protein-protein interactions as exemplified by PU.1 (Zhao et al., 2022) and the GR (Yang-Yen et al., 1990). However, it is the interaction of AP-1 with co-factors that is the key to linking extrinsic signals to the binding of these factors to enhancer activity. Seminal work by Gordon Hager studied the interaction of AP-1 with the GR and provided a first insight into how the binding of AP-1 facilitates the binding of other factors to chromatin. The GR is unable to bind to nucleosomal DNA by itself, requiring help from AP-1 factors recruiting SWI/SNF chromatin remodeling complexes. AP1 may even act in a transient fashion, a process defined as “assisted loading” (Biddie et al., 2011; Voss et al., 2011). As it turns out, the majority of AP-1/GR bound sites are occupied by nucleosomes which would support a model of enforced cooperativity in the binding of these two factors (He et al., 2013). It was also shown that AP-1 is a major interactor of the SWI/SNF complex and is required to recruit it to chromatin (Wolf et al., 2023). An elegant study by Vierbuchen et al. (2017) used fibroblasts from hybrid mouse strains to home in on regions where the cooperative binding of AP-1 with other factors is abolished by a sequence variation in individual alleles. Again, at these elements AP-1 recruits the BAF complex which facilitates tissue-specific TF binding and has been shown to be required for enhancer function (Alver et al., 2017). AP-1 is not only required for enhancer activity but also plays an important role in regulating signaling-dependent changes the 3D nuclear structure (Phanstiel et al., 2017). Given the sheer number of proper enhancers with AP-1 binding motifs in the genome, it does not come as a surprise that overexpression of JUN is incompatible with maintaining pluripotency and presents a barrier to cellular reprogramming as the fine balance of changes in TF complexes is disturbed and inappropriate enhancers are activated (Liu et al., 2015). In summary, these and our studies show that (i) enhancer activity regulated by signaling-dependent AP-1 binding is wide-spread and (ii) that signaling dependent AP-1 activity is one of the major driving forces in defining and changing cellular identities via chromatin reorganization and differential gene expression.

As described above, chronic AP-1 activation is a hallmark of AML and other cancers and drives malignant growth. A final aspect of AP-1 mediated transcriptional mechanisms is how its activity is being turned off. FOS expression is mainly regulated transcriptionally, via a serum-response element binding the TF ELK (Cavigelli et al., 1995) and by responding to MAPK signaling with an increase in Polymerase II initiation (burst) frequency (Senecal et al., 2014). JUN Kinase phosphorylates 4 sites in the TA-domain in JUN, but with different kinetics, whereby the phosphorylation of the first two sites recruits co-activators and subsequent phosphorylation shifts the balance towards co-repressor recruitment (Waudby et al., 2022). A similar mechanism has also been observed with ELK (Mylona et al., 2016) thus turning the activation of immediate-early genes and their products into a carefully orchestrated dance between activation and repression, thus fine-tuning gene expression control in response to the cellular environment. In this respect it is interesting to note that the cell-cycle repressor p57/KIP forms a complex with JUN that interferes with co-repressor recruitment, thus leading to an increase in AP-1 transcriptional stimulatory activity at the end of each cell cycle, thus preparing for the next one (Kullmann et al., 2021).

It should be noted that signaling does not just regulate inducible TFs. The list of TFs that are subject to post-transcriptional modification continues to grow, whereby such modifications often influence protein stability and co-factor recruitment. RUNX1 phosphorylation at multiple sites in response to external stimuli is a prime example (Thomson et al., 1999; Berger, 2010). Another example is GATA1 whose activity and protein stability are altered by acetylation and phosphorylation in response to EPO signaling (Boyes et al., 1998; Hernandez-Hernandez et al., 2006; Zhao et al., 2006). Acetylation influences the interaction of GATA1 with the Bromo-domain protein BRD3 which stabilizes the binding of GATA1 protein complexes in chromatin (Lamonica et al., 2011; Roe et al., 2015). Moreover, signaling molecules can also modify chromatin (histone and non-histone proteins) directly, thus regulating the expression of signaling responsive genes including JUN and FOS (Mahadevan et al., 1991; Pokholok et al., 2006; Pogna et al., 2010). The genomic response to cellular signaling therefore involves the formation of interacting, dynamic regulatory complexes in chromatin that facilitate gene expression and drive differentiation and proliferation, again demonstrating the direct impact of the signaling environment on cellular identity.

ESCs expressing an inducible dominant negative FOS (dnFOS), generated in Obier et al., 2016 (Obier et al., 2016), were differentiated for 2d to floating hematopoietic progenitors with or without Doxycycline induction and materials were processed exactly as described in (Obier et al., 2016). Cells were crosslinked with Di (N-succinimidyl) glutarate (DSG) and formaldehyde. ChIP on double-crosslinked chromatin was performed using an anti-RUNX1 antibody (Abcam #23980). Sequencing libraries were prepared using the KAPA Hyper Prep Kit as exactly described in (Obier et al., 2016).

TEAD4 dnFOS ChIP-Seq data from (Obier et al., 2016) was downloaded from GSE79320. RUNX1 dnFOS ChIP data was generated as described above. For TEAD4 and RUNX1 dnFOS data raw sequencing reads were trimmed using Trimmomatic version 0.39 (Bolger et al., 2014) to remove low quality sequences and adaptors. Reads were then aligned to the mouse 10 mm genome using Bowtie2 version 2.3.5. (Langmead and Salzberg, 2012). using the parameters–very-sensitive-local. PCR duplicates were removed using Picard tools version 2.20.2 using the MarkDuplicates function (https://broadinstitute.github.io/picard). Peaks were called using MACS2 using the options -q 0.05 –keep-dup all -B--trackline. Peaks were then filtered against the 10 mm blacklist and 10 mm repeat list. For each comparison a peak union was then formed and peaks were extended 200 base pairs around the peak summit and then merged using the merge function in bedtools version 2.27.1 (Quinlan and Hall, 2010) AnnotatePeaks.pl from Homer version 4.11 was used to calculate the average tag count in peak regions with the options -size 200 -bedgraph using the bedgraphs returned from MACS2. Peaks were then filtered against DNase1 data from (Goode et al., 2016). Tag counts were then normalized in R version 4.4.1 as tag-count per million and a peak was taken as being differentially accessible if it had a 2-fold increase or decrease between the two samples before and after Dox induction. Tag-density plots were created by first ranking peaks by fold-difference. The read density was determined using the annotatePeaks.pl function in Homer version 4.11 (Heinz et al., 2010) using the options -size 2000 -hist 10 -ghist -bedgraph. The output was then plotted as a heatmap using Java TreeView version 1.1.6r4 (Saldanha, 2004). Fold-change analysis centred on RUNX1 peaks was performed as previously described (Obier et al., 2016)

Homer was used to perform a de-novo motif search on differentially bound peak sets using the findMotifGenome.pl function. The sites of enriched motifs were determined using the annotatePeaks.pl function in Homer using the options -size 2,000 -hist 10 -ghist -m and plotted using Java TreeView.

Sites which were differentially bound by TEAD4 or RUNX1 before and after dnFOS induction were intersected with functionally validated enhancers from to (Edginton-White et al., 2023) to generate lists of enhancers bound by TEAD4 and RUNX1. TEAD4, RUNX1 and FOS differentially bound sites were also intersected with VEGF responsive enhancers using bedtools intersect. Venn diagrams were made using BioVenn (Hulsen et al., 2008)